Automated document verification using computer vision

01. The Challenge

A large financial services company, Lazard, faced significant challenges in its document verification process. The traditional manual verification method was time-consuming, error-prone, and required substantial human resources. This inefficiency led to delays in customer onboarding and loan processing, ultimately affecting customer satisfaction and operational costs.

Key challenges included:

1. High Volume of Documents: Processing thousands of documents daily, including IDs, bank statements, and contracts.
2. Human Error: Manual verification was susceptible to errors, leading to potential fraud risks and compliance issues.
3. Operational Bottlenecks: Long processing times delayed customer onboarding and service delivery.
4. Compliance and Security: Ensuring that the document verification process met strict regulatory requirements and maintained high security standards.

02. The Solution

Lazard partnered with our AI agency to implement an Automated Document Verification system using Computer Vision technology, leveraging the Document Text Recognition (DocTR) framework. Our solution included the following components:

1. DocTR Integration: Implemented DocTR to extract and validate text from various document types, ensuring high accuracy and speed.‍
2. Automated Workflow: Developed an end-to-end automated workflow for document submission, verification, and approval.‍
3. Real-Time Processing: Enabled real-time document verification, reducing processing time from days to minutes.‍
4. Fraud Detection: Implemented advanced algorithms to detect anomalies and potential fraud in submitted documents.

03. The Result

The integration of Automated Document Verification with DocTR resulted in significant improvements for Lazard:

1. Increased Efficiency: Reduced document processing time by 85%, enabling faster customer onboarding and loan approvals.‍
2. Enhanced Accuracy: Achieved a 99% accuracy rate in document verification, significantly reducing errors and fraud risks.‍
3. Cost Savings: Decreased operational costs by 60% through reduced manual labor and faster processing times.‍
4. Improved Customer Satisfaction: Enhanced customer experience with quicker turnaround times and reliable verification processes.‍
5. Scalability: The solution was scalable, allowing Lazard to handle growing document volumes without compromising on speed or accuracy.

"For document classification, we employed the YOLO model. For information extraction, we initially used Tesseract but later transitioned to DocTR. This shift was driven by DocTR's superior ability to accurately extract information from images of highly variable quality" - Andrzej Nowak, CTO at Aidentico

‍

‍